Fine-Tuning Universal Machine Learning Potentials Enables Efficient Computation of Li Electrode Potentials in Battery Electrolytes

Abstract The electrode potential of Li metal is fundamental to the performance and stability of Li-metal batteries, yet its accurate prediction across diverse electrolyte chemistries remains a challenge. Standard quantum chemistry using cluster models fails to account for explicit bulk solvation, while ab initio molecular dynamics is restricted by the high computational cost of thermodynamic sampling. In this work, we present an integrated framework that combines a universal machine learning potential for electrolytes with a data-efficient fine-tuning strategy to compute Li electrode potentials. This workflow achieves ab initio level accuracy using ∼200 extra density functional theory calculations per electrolyte system. Furthermore, the identified relationship between local Li+ solvation environments and electrode potential shifts provides mechanistic insights into electrolyte-dependent redox behavior and offers guidance for rational electrolyte design.